Quick Install windows system of detailed tutorial pytorch

Pip of difference and conda

Conda has been used before and pip, sometimes often mix the two. But today I discovered two things are not installed in one place, so find some things their own equipment, but found that the package can not always recognize, in particular, have been wondering when the operating environment, notice the problem until today so to sum up the difference between the two.

pip

pip专门管理Python包
编译源码中的所有内容。 （源码安装）
由核心Python社区所支持（即，Python 3.4+包含可自动增强pip的代码）。

conda

Python不可知论者。 现有软件包的主要重点是Python，而conda本身是用Python编写的，但你也可以为
C库或R软件包或任何其他软件包提供conda软件包。

安装二进制文件。 有一个名为conda build的工具，它可以从源代码构建软件包，但conda install本身
会安装已经构建的conda软件包中的东西

外部: Conda是Anaconda的包管理器，由Continuum Analytics提供的Python发行版，但它也可以在
Anaconda之外使用。 您可以使用现有的Python安装，通过pip安装它（尽管除非您有充分理由使用现有安
装，否则不建议这样做）。

Tips

conda installation is best to build another virtual environment, do not install the base in.

View cuda version count force and the corresponding graphics

View version of the card count corresponding force

https://developer.nvidia.com/cuda-gpus

View cuda corresponding version of the card

1. Find nvcuda64.dll computer graphics in their own folder
2. Details> - - Right Properties> Name
3. Cuda can be found in the corresponding version of a map: My graphics support is 8.0
4. Quguan network https://pytorch.org find the corresponding pytorch version corresponds to Figure II
5. Note that we must find duda version corresponding to the current page is not displayed if you need later to clicking on the left: previous versions of pytorch
6. Find the corresponding version of conda installation command such as mine is: conda install pytorch == 1.0.0 torchvision == 0.2.1 cuda80 -c pytorch correspond to Figure 3: Remember to get back with

Figure I

Figure II

Figure III

How to speed up the installation

Tsinghua original image

1. Tsinghua original image with a simple sentence: the real installation on two commands, Tsinghua handover original image, the installation
2. Open cmd
3. conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch/
   
4. conda install pytorch == 1.0.0 torchvision == 0.2.1 cuda80 (note found in the above installation command to remove the -c pytorch)

Simple check that the installation was successful

直接上图：